Does Claude use Brave Search?

The evidence strongly suggests yes, as of mid-2026. Anthropic added Brave Search to its public subprocessor list in March 2025, independent testers found a 'BraveSearchParams' setting inside Claude's web-search tool, and Profound's mid-2026 analysis reported that 79.2% of Claude's citations matched Brave's top-ten results for the same query. Anthropic has not formally documented this, so treat it as well-corroborated but not officially confirmed — and expect it to change.

How is getting cited in Claude different from ranking on Google?

Claude's web search appears to retrieve from Brave's independent index, not Google's — so Google rank does not carry over the way it does in Google AI Mode or AI Overviews. A page can rank well on Google and never surface in Claude if it is weak or absent in Brave's index. The practical implication: being reachable and well-represented in Brave matters as much as your Google position.

What does Claude quote when it cites a page?

Claude's Citations API returns a short cited_text span — roughly 150 characters in practice — tied to the source URL, so it lifts a clean, self-contained passage rather than a whole page. That rewards answer-first writing: a direct, quotable sentence near the top of each section is far more liftable than the same point buried mid-paragraph.

How to get cited in Claude (and why it leans on Brave Search)

Claude answers current questions by searching the live web, synthesising what it finds, and showing citations to the pages it used. The twist that sets it apart from the other engines: as of mid-2026, that search appears to be powered by Brave Search's independent index — not Google's, not Bing's. This is part 7 of the engine guides, and getting cited in Claude means understanding both how Claude retrieves and how Brave indexes.

Does Claude really use Brave Search?

The evidence is strong but unofficial. Three independent signals point the same way:

The subprocessor list. Anthropic added "Brave Search" to its public subprocessor list in March 2025 — the partners that process Claude data. Programmer Simon Willison and TechCrunch reported the addition at the time, alongside identical citations appearing in Claude and Brave for the same query.
The internal parameter. Testers found a setting named BraveSearchParams inside Claude's own web-search tool — a strong hint about what supplies the results.
The citation overlap. Profound (Josh Blyskal, mid-2026) reported that 79.2% of Claude's citations came directly from Brave's top-ten results for the equivalent search, with a measured 86.7% overlap (13 of 15 results) and "no meaningful reranking." No other major model tracks its search provider this closely.

Treat this as well-corroborated but not officially confirmed by Anthropic, and expect the arrangement to evolve. The Profound figures are single-vendor and self-reported — directionally credible, not gospel. The durable takeaway survives the caveats: Claude's live answers lean on an index that is not Google's.

How does Claude's web search work?

Claude blends two sources, much like ChatGPT:

Trained knowledge. Anthropic's training crawler, ClaudeBot, reads the open web; what it learns becomes Claude's baked-in picture of your brand. This has no citations and updates slowly, on model releases.
Live web search. For current or comparative questions, Claude searches the web — apparently via Brave — and grounds the answer in retrieved pages. These answers carry citations through Anthropic's Citations API, which returns a short cited_text span (about 150 characters in practice) tied to each source URL.

A single question is decomposed through query fan-out into several sub-queries, each retrieving its own sources, then synthesised into one grounded answer. So one buying question becomes several retrievals — and several chances to be cited.

Claude's web search appears to read from Brave's index, not Google's. If your Google rank is strong but Brave barely knows you, Claude may never see the page Google loves.

What gets cited in Claude?

Claude narrows many candidate passages to the handful it quotes. The selectors are the familiar five quality bars, but filtered through Brave's retrieval first:

Selector	What it means for your page
Present in Brave	Reachable and indexed by Brave's independent crawl — not just Google
Extractable	Answer-first passages a 150-char citation can lift cleanly
Specific	Numeric, named, dated claims; vague marketing copy gets skipped
Structured	Schema, tables, and clean markup Brave extracts well (see below)
Corroborated	Agreed by multiple credible sources, not a lone claim

For "best X" and comparison questions, Claude — like every engine — leans on independent roundups and community threads, so earned placement in the lists it cites matters alongside your own pages.

How do you rank in Brave's index?

Brave publishes no ranking guidelines, so nobody optimises Brave directly. But Brave is unusually open about how it reads pages, which tells you what to give it. Per Brave's own Search API documentation (mid-2026):

It is a genuinely independent index — Brave reports over 30 billion pages, refreshed by "100 million page updates a day," built from its own crawl plus anonymised data contributed by Brave browser users. Being in Google's index does not put you in Brave's.
It sells an "LLM context" / AI-grounding endpoint designed to feed models clean, factual, real-time content and cut hallucinations — the exact surface a tool like Claude would consume.
It prioritises structured extraction. Brave's docs emphasise schema-enriched results — JSON-LD structured data, tables (down to row-level granularity), forum discussions, and clean query-optimised text snippets. Pages that expose facts in structured, extractable form are easier for Brave to represent well.

The practical reading: the same GEO fundamentals win here, with extra weight on being reachable to crawlers generally (don't assume Google access equals Brave access) and on clean structured data that Brave parses cleanly.

The playbook to get cited in Claude

Open the gates wide — for everyone. Allow ClaudeBot and confirm your CDN isn't silently 403-ing AI crawlers; the crawler directory has the tokens. Because Claude leans on Brave, also make sure you aren't blocking general web crawlers — a narrow allowlist that only admits Google can leave you invisible in Brave's index.
Don't assume Google coverage equals Brave coverage. Search your key terms on Brave Search directly and see whether your pages appear. If they don't, the cause is upstream of anything Claude does.
Write answer-first. Lead each section with the direct answer in 40–60 words so the ~150-character cited_text span has a clean sentence to lift.
Mark up your facts. Add JSON-LD schema, put specs and comparisons in real tables, and keep forum-style Q&A crawlable — the formats Brave's docs say it extracts best.
Date and corroborate claims. Specific, dated, source-attributed statements survive retrieval; unattributable numbers get cut.
Pursue earned coverage in the reviews and roundups Claude cites for your category — third-party sources win the "best" questions.
Keep competitive pages fresh so live retrieval keeps finding current, dated content before they fall off the citation cliff.

For a contrast, Perplexity cites from its own blended retrieval and Google AI Mode grounds in Google's index — so the same page can win in one engine and lose in another. That divergence is exactly why single-engine checks mislead.

How do you measure your Claude visibility?

Ask Claude your highest-intent customer questions, repeatedly, and track two things over time: whether your brand is named and whether your pages are cited as sources — then compare that against the same prompts in ChatGPT, Perplexity, and Google's surfaces, since Claude's Brave-grounded answers will diverge from the Google-grounded ones. That cross-engine, multi-prompt read — presence, citations, and sentiment tracked daily rather than spot-checked — is exactly what Buffy Intel is built for.